A survey of natural language interaction and environmental perception in intelligent robotics
Year: 2026
Type: article
Abstract: Natural language interaction and physical environment understanding are two core challenges for improving the autonomy of home service robots. This paper provides a systematic review of the evolution from traditional robotic methods to embodied intelligence enabled by large language models (LLMs), vision-language models (VLMs), and vision-language-action (VLA) models. For environmental understanding, we review the transition from geometric maps to semantic representations and behavioral navigation, emphasizing the roles of semantic SLAM, hierarchical scene representations, and VLMs. For natural language interaction, we analyze the limitations of rule-based approaches and the advances enabled by LLMs in task planning, reasoning, and code generation. Unlike previous surveys focusing separately on perception or language interaction, this review establishes a unified taxonomy and analyzes representative LLM/VLM/VLA-based robotic systems. We further discuss emerging trends toward multimodal reasoning, language-guided action generation, and adaptive embodied intelligence, while identifying key challenges in generalization, real-time deployment, safety, and interpretability. Finally, future directions toward reliable and adaptive embodied robotic systems are discussed.
Source: Discover Applied Sciences
Authors: Xiaoqing Shao, Juan Ding, Pengwei Zhang, Qing Zhao
Language: English
Subfield: Control and Systems Engineering
Field: Engineering
Domain: Physical Sciences
Keywords: natural language interaction, environmental perception, home service robots, large language models, vision-language models, vision-language-action models, semantic SLAM, semantic maps, hierarchical scene representation, task planning, code generation, multimodal reasoning, embodied AI, robot autonomy
Open Access status: diamond
OpenAlex ID: W7236855219
